metadata
dataset_info:
features:
- name: image
dtype:
array3_d:
shape:
- 3
- 288
- 384
dtype: float32
- name: segmentation
dtype:
array2_d:
shape:
- 288
- 384
dtype: int64
- name: depth
dtype:
array3_d:
shape:
- 1
- 288
- 384
dtype: float32
- name: normal
dtype:
array3_d:
shape:
- 3
- 288
- 384
dtype: float32
- name: noise
dtype:
array3_d:
shape:
- 1
- 288
- 384
dtype: float32
splits:
- name: train
num_bytes: 3525109500
num_examples: 795
- name: val
num_bytes: 2899901400
num_examples: 654
download_size: 2971250125
dataset_size: 6425010900
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: val
path: data/val-*
task_categories:
- depth-estimation
- image-segmentation
- image-feature-extraction
size_categories:
- 1K<n<10K
This is the NYUv2 dataset for scene understanding tasks. I downloaded the original data from the Tsinghua Cloud and transformed it into Huggingface Dataset. Credit to ForkMerge: Mitigating Negative Transfer in Auxiliary-Task Learning.
Dataset Information
This data contains two splits: 'train' and 'val' (used as test dataset).
Each sample in the dataset has 5 items: 'image', 'segmentation', 'depth', 'normal', and 'noise'.
The noise is generated using torch.rand().
Usage
dataset = load_dataset('tanganke/nyuv2')
dataset = dataset.with_format('torch') # this will convert the items into `torch.Tensor` objects
this will return a DatasetDict:
DatasetDict({
train: Dataset({
features: ['image', 'segmentation', 'depth', 'normal', 'noise'],
num_rows: 795
})
val: Dataset({
features: ['image', 'segmentation', 'depth', 'normal', 'noise'],
num_rows: 654
})
})
The features:
{'image': Array3D(shape=(3, 288, 384), dtype='float32', id=None),
'segmentation': Array2D(shape=(288, 384), dtype='int64', id=None),
'depth': Array3D(shape=(1, 288, 384), dtype='float32', id=None),
'normal': Array3D(shape=(3, 288, 384), dtype='float32', id=None),
'noise': Array3D(shape=(1, 288, 384), dtype='float32', id=None)}